|
IBM's eServer i5 Plans for 2005 and Beyond
by Timothy Prickett Morgan
For the past couple of years, the engineers and product managers at IBM's Rochester labs have sat down with me and talked about the big picture in the iSeries market in the coming years. While IBM can't give a lot of specifics on near-term announcements, the company is always willing to talk about, in a very general sense, what is coming down the pike in iSeries land.
IBM won't be making any major hardware and software announcements with the iSeries line in 2005, say my sources at Big Blue. IBM began the transformation of the iSeries into the eServer i5 in January 2003, when it repackaged and repriced the iSeries hardware, its OS/400 operating system, and the 5250 interactive processing features. In 2004, the transition was completed by bringing iSeries and pSeries hardware pricing to parity and breaking out interactive processing capacity as a separately priced software feature (which it really is) rather than a hardware feature (which it never was, even though IBM said it was for close to a decade). If 2004 was the year of re-engineering the iSeries, then 2005 appears to be the year for tinkering. That's not to say that there won't be significant announcements; they just won't be as broad and as deep as the eServer i5 and i5/OS announcements this year.
On the hardware front, IBM apparently has no plans to make i5 variants of the just-announced eight-way p5 575, which crams four dual-core Power5 chips into a 2U rack-mounted server. This is the density computing performance available in the world today, and IBM has created it explicitly to go after high-performance and technical computing markets, where the flops per watt and the flops per rack are the main factors in determining which server architecture gets chosen for a big cluster. IBM is also readying a 1U, two-way p5 and OpenPower (the Linux-only variant of the Power5 "Squadron" family) server for early 2005, presumably to be called the p5 525 when it runs AIX and Linux and the OpenPower 725 when it runs only Linux. IBM Rochester has no plans to support this server, either, mainly because such densities are apparently less important in the iSeries world than in the HPC markets.
What is important to i5 buyers is expandability, particularly at the low-end, where customers typically want a server that has room for an adequate amount of disk capacity and peripherals for use today and room to grow for later. According to surveys done by IBM, the typical OS/400 shop keeps its AS/400, iSeries, and i5 servers around for twice as long as the typical X86 shop keeps around a Pentium or Xeon server. So on machines like the Power4-based iSeries Model 825 or the Power5-based i5 Model 550 (the belly of the iSeries and i5 product lines), IBM has enough expansion room to add up to 10 TB of disk capacity; some customers with these machines are already pushing 3 or 4 TB of disk. The Sarbanes-Oxley Act and other compliance regulations have forced companies to stockpile more of their electronic information. A few years ago, disk capacity on the iSeries was prohibitively expensive, so customers tended to have relatively skinny configurations. But IBM has cut the cost of disk capacity on the i5 servers, and it has discovered an old law of economics called elasticity of demand: as the price of a component drops, consumption increases, often by so much that you can make a lot more money. (It's also called making it up in volume.)
That's why, in 2005, IBM will continue to enhance the storage on the iSeries. During the first half of 2005, IBM plans to make some I/O enhancements to the i5 line. Exactly what they will be is unclear, but fatter and faster disk drives are always a good guess. There's talk that IBM will more tightly integrate the i5 servers with the new DS6000 midrange SAN arrays that the company launched this summer, too. IBM will probably not move to 2.5-inch SCSI or ATA disk drives in 2005, but the word on the street is that IBM will move from PCI-X buses and peripheral cards up to the faster PCI Express buses and peripherals, most likely getting it done in late 2005.
Those I/O enhancements will coincide, very roughly, with an expected bump in the processor performance of the i5 line during the second half of 2005. The eServer i5 machines use the 1.5 GHz and 1.65 GHz variants of the Power5 processors. IBM is already shipping 1.9 GHz Power5s in its eServer p5 line, and will probably give customers the option of using these 1.9 GHz chips in the i5s, too. However, IBM is expected to roll out the Power5+ processors, which move from the 130 nanometer, copper/SOI process used in the Power4+ and Power5 chips to a 90 nanometer copper/SOI process, sometime before the end of 2005 as well. These Power5+ chips could come out at similar or slightly higher frequencies and consume a lot less power, or IBM could hold the power consumption more or less constant and boost the clock speed to 2.5 GHz or a little higher (maybe even 3 GHz) on the first pass with this new process.
IBM probably could hit 4 GHz with the Power5+ chip using the 90 nanometer process and not even bother going to a Power6 chip. The question now is whether it will push clocks that hard. No vendor can even touch the Power5 chip when it comes to raw performance, so IBM might start playing the good guy on electric consumption and heat dissipation, perhaps even going so far as to bifurcate the Power5 line, offering low-power variants that run at lower voltages and clock speeds for embedded, dense, or blade computing environments, while offering the full-power variants with high clock speeds for those applications where peak performance is a must. None of my sources at IBM has said that this bifurcation is in the works, but it is a good idea and a potentially differentiating one.
IBM is in development of the Power6 systems, and it will be using the 90 nanometer and future 65 nanometer processes it is creating with many partners to radically improve the performance of its Power chips. Exactly how much is unclear, but IBM sources say that the jump from Power5 to Power6 will involve a big jump in gigahertz--more than the jump from Power4 to Power5. The fastest initial Power4 clocked at 1.3 GHz, and the fastest Power5 clocks at 1.9 GHz, which is a bump of 46 percent. It seems likely that Power6 chips will probably start out at 3 GHz and push up to 4 GHz.
My sources say that it is not a foregone conclusion that Power6 chips will have more than two cores per chip. With that 90 nanometer process, IBM can undoubtedly put four whole Power cores--and probably a lot of other electronics--on a chip. Those sources say that the problem with a four-core Power6--and particularly one running at a high clock speed--is that is can end up being starved by relatively slow cache and main memory subsystems. What IBM probably should do is rethink this whole clock speed game for Power6, as Intel has been forced to do. IBM needs to get more of the system components onto the chip. The Power4 put two cores with their own L1 caches, the L1 cache controllers, a shared L2 cache, and a single L2 cache controller onto the chip and put the L3 cache off the die. With Power5, IBM added simultaneous multithreading (SMT), boosted the size of the L2 cache, moved the L3 cache controller into the chip, and moved the L3 cache into the chip package. If IBM can do it, it seems likely that Power6 will include an on-die L3 cache and an integrated L4 controller, at least for the high-end server variants of the chip. IBM will reportedly also add a lot more functions for self-management from the microcode underpinning OS/400 and AIX into the chip itself. It would not be surprising for the large pieces of the virtualization embodied in the Virtualization Engine to somehow be implemented in chip transistors and firmware loaded into the processor.
Given all of these possible paths for Power6 chips, it is hard to reckon what the performance boost might be with the Power6 systems. IBM could easily double or triple performance over the first-generation Power5 eServer i5s. But we all know that no one needs IBM to do that. Quite frankly, performance is not an issue with the i5. Price is.
On the OS/400 software front, IBM sources say that there is not going to be a lot of action with i5/OS in the coming year, either. After input from customers in the past few years, IBM has decided to roll out a new OS/400 release every 18 months or so and to support releases for a longer period of time. Perhaps most significant, Power5 servers require i5/OS V5R3, but Power6 servers will also run i5/OS V5R3, as well as the future i6/OS V5R4 (or i6/OS V6R1, if IBM calls it that). This is called N-1 support in the industry. Since the beginning of the AS/400 line 16 years ago, moving from one generation of servers to the next has always meant upgrading a release or jumping a version of the operating system. Now that IBM is coding OS/400 to the hypervisor layer, it can offer N-1 support on future i5 iron.
Next year, IBM will be focusing on updates to WebSphere, Domino, Workplace, and other middleware software that rides on top of i5/OS, and working to better integrate that software with the operating system. IBM is not planning on supporting Solaris, Mac OS X, or any other operating system on the i5 platform, but my sources say that it could expand the number of special i5/OS editions that are sold. IBM could make more Solution Edition and Domino Edition machines, which offer discounts on i5 hardware and i5/OS software for machines dedicated to specific tasks (in this case, running specific ERP suites or Domino groupware).
Articles on the Future of the iSeries from Prior Years
|
MAXIMUM AVAILABILITY
